The representation of correlated clutter textures in coherent images
Christopher J. Oliver · Inverse Problems · 1988
Some of the fundamental problems of representing correlated clutter textures in coherent images are considered. A general class of noise models, derived from gaussian random walks, has been demonstrated to have statistical properties consistent with a variety of examples of coherent clutter. The author examines further the extent to which such a noise model can represent clutter textures. The noise model entails two basic assumptions; (i) that the textures are homogeneous and (ii) that the textures can be completely represented by the average two-point statistics, i.e. the autocorrelation function (ACF). The author defines quantitative measures to test texture similarity in terms of the intensity ACF and the single-point statistics. Using these tests the author demonstrates that textures can be simulated with the same average two-point statistics. By visual comparison of such simulated textures with original images, the author demonstrates that many textures are, indeed, well represented by the noise model. In other examples the observed discrepancy can be attributed to specific failures of the two assumptions of the underlying noise model.